The purposes of this research were (1) to describe the learning quality of PBL through dyadic interaction approach to mathematical reasoning ability of grade X students of IT Al Irsyad Purwokerto High School (2) to find pattern of students reasoning ability based on mathematical belief. This study uses a mixed method. Data analysis started from the analysis of test items. The analysis uses the prerequisite test and then hypothesis testing uses rara average (t-test), then the proportion of comparative tests (test-z) is to calculate the classical completeness. Further testing of determining the difference between the two classes uses different test average (t-test right side). Qualitative data analysis uses qualitative description. The results of quantitative research shows that learning class with PBL and dyadic interaction approach reached classical total 80%. The average difference test showed class’ results with PBL dyadic interaction approach better than a class activity with PBL. Subjects with very low belief can only fulfill 1 reasoning indicator well. Subjects with low belief can fulfill 3 reasoning indicators well and have not been able to fulfill 1 other indicator. Subjects with high belief can fulfill 4 indicators where 1 indicator is imperfect and subject with very high belief can fulfill all the indicators of reasoning well and complete.